In the information processing system of the prior art, information has been stored to storage devices, hard disk drive for one, which is directly connected in direct to a computer system. Information stored in the storage devices may be accessed only from the computers directly connected to the storage devices. Therefore, to access the information from the other computer systems, it has been required to access via the computer directly connected to the storage devices.
However, the progress of a network technology and rapidly increasing in amount of information to be stored in recent years have caused to stimulate to separate a storage system to store information from a computer system to process information. As a result, a new storage system architecture has been developed. This storage system can realize sharing information among a plurality of computer systems through a network connection. A storage system to be connected to such network is referred to as a network storage.
As examples of such network storage, there have been proposed a SAN (Storage Area Network) storage which can provide a block access through the SAN connection, a NAS (Network Attached Storage) which can provide a file access through the IP network or the Infiniband connection, and a Web storage which can provide access profiles of a HTTP protocol used as an interface of Web access and an expanded protocol of the HTTP protocol.
With spread of network storage, a system manager is required to introduce management of a plurality of network storage connected to a network. For example, in the case where amount of data to be stored exceeds the maximum capacity of the existing network storage, it is required to add new network storages to reconfigure the system for reallocation of data or the like. Rapid growth of demand for storage capacity has imposed repeated reconfigurations of system and expanded system management cost.
In order to reduce system management cost, it is essential to introduce a storage virtualization technology with which a plurality of network storages are formed as a virtually single network storage from the computer system side and thereby any adverse effect like adding new network storages, is not spread to the system as a whole. Various systems have been developed and proposed in regard to such a storage virtualization technology.
For instance, the Zebra (refer to a non patent reference 1) presented by Hartman et. al. in 1995 proposes a system for storing a single file into a plurality of distributed network storages by striping this file. In this system, a file is divided in every constant length and sequentially stored into the network storages in round-robin fashion. In the Zebra reference, a server is provided for centralized management to know in what sequence files are stored in the distributed network storages. Therefore, this reference is characterized in that the data of file itself is accessed in the file access process after an inquiry is issued for information regarding the storing position of file to such management server. Namely, in the Zebra reference, a plurality of network storages are virtually formed as a large network storage by storing files into a plurality of distributed servers and centrally managing allocations of the files with a single server.
Moreover, a patent reference 1 by Alexander H. Frey et. al. proposes a system for striping file with non-centralized management servers system which are different from the centralized management server system of the Zebra reference. In this system, the striping information which has been in the centralized management server with the Zebra reference is stored for a distributed management in a plurality of distributed network storages by embedding starter node information which is storing the striping information of files into a file identifier. When a client wants to access to files, the client accesses first to the starter node to determine the storing position of files and transfers, if necessary, an access request to the server which is storing file blocks to be accessed because the starter node information is stored in the file identifier. Therefore, an access to the relevant file from a client can be realized only by issuing a request to the starter node, resulting in the merit that it is not requested to consider the storing position of files through the striping thereof.
Moreover, in the “distributed file management apparatus and distributed file management system” of the patent reference 2, status management sections are provided in a plurality of server computers for management of load by holding load information, while partial files of distributed files are identified by storing partial files forming a part or the entire part of the distributed files to a storage device and thereby a server computer for processing partial files based on the load information stored in the status management sections is determined. Accordingly, centralization of load to the particular server can be prevented.